Sanders ASI Ban Bill Draws Fire from Safety Researchers and Sacks
Sen. Sanders and Rep. Casar introduced the 2026 Ban Artificial Superintelligence Act, which would create a federal Department of AI (DoAI) and pause training of models above 10^25 ops until the department is ready. Harvard researcher Stephen Casper posted a long thread analyzing the bill section by section, concluding that its boldness may be one of humanity's best chances to avert catastrophic AI risk, though its design is flawed with vague, hard-to-enforce definitions.
Confirmed
- The bill would establish a federal DoAI and pause training of models exceeding 10^25 ops until it is operational.
- Models above 10^25 ops face licensing-related requirements, per Casper's breakdown.
- Former White House AI lead David Sacks criticized the bill's overly broad definition of "superintelligence," arguing today's models might already qualify and open-source models would be caught; he also revealed the bill would impose up to 20 years in prison for developers.
Why it matters
- The bill is among the most aggressive attempts in Congress to regulate frontier AI, and if advanced it would directly affect large-model training and the open-source ecosystem.
- Casper and Sacks, from AI-safety and industry/policy perspectives respectively, offer sharply divergent assessments, exposing the core tension between decisive risk prevention and overbroad definitions.
- Casper stated he voices his true views even outside the Overton window, indicating his support for decisive regulation is not performative positioning.
2026-09-26 ~ 2026-09-27 · 5 related posts
- Episode 1: Sanders Bill Would Permanently Ban Superintelligence, With Up to 20 Years in Prison(2026-09-04, 21 posts)
- Episode 2: Sanders: Rogue Superintelligence Is All Humanity's Problem(2026-09-04, 3 posts)
- Episode 3: UK Parliament Sees World's First Anti-Superintelligence Bill as US Follows Suit(2026-09-08, 8 posts)
- Episode 4: Sanders Moves to Ban Superintelligence, Sparking AI Debate(2026-09-09, 3 posts)
- Episode 5: 71 UK Lawmakers Back Bill to Ban Superintelligent AI(2026-09-12, 10 posts)
- Episode 6: Viral Claim: Sanders AI Bill Would Jail Frontier AI Developers Up to 20 Years(2026-09-12, 12 posts)
- Episode 7: Sanders Proposes Bill to Ban Superintelligence and Pause Advanced AI(2026-09-12, 2 posts)
- Episode 8: Sanders Proposes 20-Year Prison Terms for ASI Developers(2026-09-14, 2 posts)
- Episode 9: Jensen Huang on Ezra Klein Show: Shut down labs that can't control their models(2026-09-23, 27 posts)
- Episode 10: Sanders Bill Proposes Permanent Ban on Superintelligence, Up to 20 Years in Prison(2026-09-23, 23 posts)
- Episode 11: Sanders ASI Ban Bill Draws Fire from Safety Researchers and Sacks(2026-09-26, 5 posts)
Primary sources
- Sanders/Casar 'Ban ASI Act of 2026' proposes federal AI dept and training pause above 10^25 ops — StephenLCasper · 2026-09-26
- [source] Harvard researcher dissects Sanders' ASI ban bill: too vague, ignores hardware supply chain — StephenLCasper · 2026-09-26
- AI safety researcher: I'll share views outside the Overton window, not popularity plays — StephenLCasper · 2026-09-26
- [source] Sanders AI bill could bring 20-year sentences, its superintelligence bar may already be met by open models — victor_explore · 2026-09-27
1 near-duplicate retellings: StephenLCasper